Here is a fact about modern medicine that turns out to be inconvenient. The most precise drugs we have - antibodies, the biologics that hit a single target and leave the rest of the body alone - almost all have to be injected or infused. They are large, they are fragile, and the human gut is very good at destroying large, fragile things. So the most surgical medicines arrive by the least convenient route.
On the other end sit small molecules: cheap, oral, easy to make, and able to slip inside cells. Their problem is the opposite. They are small enough to go anywhere but often too small to grip the flat, sprawling surfaces where two proteins meet. Those protein-protein interactions are where a lot of disease lives, and the industry spent years calling them "undruggable," which in drug discovery is less a scientific claim than an admission of defeat.
Insamo is one of a handful of companies betting the escape hatch is a molecule in the middle: the macrocyclic peptide. A ring of amino acids big enough to grab a protein like an antibody does, engineered to be stable and greasy enough to cross a cell membrane - and, ideally, sturdy enough to be taken as a pill. It is a genuinely hard chemistry problem. Membrane permeability and antibody-like binding tend to pull in opposite directions, and finding the rare molecule that does both by hand is slow, expensive work.
Insamo's answer is to stop doing it by hand. The company - spun out of UC Berkeley and now housed at the university's Bakar Labs incubator - has built a platform that runs the search as an automated loop: design a batch of candidates with machine learning, synthesize them with parallel chemistry, test them with high-throughput molecular biology, then feed the results back into the model and go again. Each turn of the loop produces data. Enough turns, the company says, produces trillions of proprietary experimental data points - the kind of dataset that only exists if you generate it yourself.
That is the quiet insight underneath the science. The competitive advantage is not really the model; anyone can fine-tune a model. It is the data, and the only way to get this particular data is to run the experiments. Insamo is building a wet lab that behaves a bit like a search engine, indexing chemical space one measured molecule at a time.
